Our next MEAS Department Seminar is this coming Monday, October 24, 330PM, available only with Zoom.
Speaker – Maria J. Molina, Assistant Professor, University of Maryland, College Park (hosted by G. Lackmann)
Seminar Title – Machine Learning Applications for Extending Earth System Prediction
Bio – Dr. Maria J. Molina is an Assistant Professor within the Department of Atmospheric and Oceanic Science at the University of Maryland, College Park. She runs a research group that focuses on data science for climate and extremes. Maria is also affiliated with the National Center for Atmospheric Research in Boulder, Colorado and is an Adjunct Assistant Professor within the Department of Marine, Earth, and Atmospheric Sciences at North Carolina State University. Maria also serves as Co-Chair of the AMS Early Career Leadership Academy, is a member of the AMS Board on Representation, Accessibility, Inclusion, and Diversity (BRAID), and as an Academia Ambassador for the AMS Committee for Hispanic and Latinx Advancement (CHALA). https://mariajmolina.github.io/
Abstract – Chaos theory, which is the concept of sensitivity to initial state for numerical weather prediction, is at the root of why the skillful deterministic prediction of weather at the subseasonal-to-seasonal (S2S) timescales using state-of-the-art forecasting models remains extremely challenging. Predictability stemming from atmospheric initial conditions is also substantially reduced beyond approximately two weeks and the ocean generally does not offer added predictability until a trajectory reaches the seasonal timescale. These challenges motivate the use of machine learning methods for S2S prediction. Two approaches for S2S prediction will be highlighted: (1) an unsupervised learning approach and (2) a supervised learning approach. The first study focuses on assessing the representation and predictability of North American weather regimes, which are persistent large-scale atmospheric patterns. The second study focuses on the use of a deep learning model (specifically a U-Net) for bias correction of S2S forecasts of global temperature and precipitation. We also highlight several forecasts that exhibit high predictability at the subseasonal time scale, seemingly “defying chaos theory,” along with the potential reasons for skill.